View
8
Download
0
Category
Preview:
Citation preview
Reproducible Research.. Using Sweave, Knitr and Pandoc
Aedín Culhane aedin@jimmy.harvard.edu
Nov 20th 2012
My R Course Website http://bcb.dfci.harvard.edu/~aedin/
My HSPH homepage http://www.hsph.harvard.edu/research/aedin-culhane/
When issues of reproducibility arise
• ``Remember that microarray analysis you did six months ago? We ran a few more arrays. Can you add them to the project and repeat the same analysis?''
• ``The statistical analyst who looked at the data I generated previously is no longer available. Can you get someone else to analyze my new data set using the same methods (and thus producing a report I can expect to understand)?''
• ``Please write/edit the methods sections for the abstract/paper/grant proposal I am submitting based on the analysis you did several months ago.''
From Keith Baggerly
• Selected articles published in Nature Genetics between January 2005 and December 2006 that had used profiling with microarrays
• Of the 56 items retrieved electronically, 20 articles were considered potentially eligible for the project
• The four teams were from – University of Alabama at Birmingham (UAB) – Stanford/Dana-Farber (SD) – London (L) and Ioannina/Trento (IT)
• Each team was comprised of 3-6 scientists who worked together to
evaluate each article.
Results
• Result could be reproduced n=2 • Reproduced with discrepancy n=6 • Could not be reproduced n=10
– No data n=4 (no data n=2, subset n=1, no reporter data n=1)
– Confusion over matching of data to analysis (n=2) – Specialized software required and not available
(n=1)m – Raw data available but could not be processed n=2
Reproducibility of Analysis
Ioannidis JP, Allison DB, Ball CA, Coulibaly I, Cui X, Culhane AC, et al, (2009) Repeatability of published microarray gene expression Analyses. Nature Genetics 41(2):149
Reproducible Research in R
• Sweave • Knitr • Knitr + pandoc
Typical LATEX
http://en.wikibooks.org/wiki/LaTeX/simple.tex
Blah blah blah blah…..
Sweave
• R embedded in Latex • Produce pdf or html files • R code is run each time, so you are sure the
code works • Document includes results of the code Sweave (filename.rnw) Stangle (filename.rnw)
Quick Start to Sweave
• Insert an R code chunk starting with << >>= • Terminate the R code chunk with an @ sign
<<easySweave>>= x <- mean(1:10) print(x) @
• Save LaTeX with extension ``Rnw''
Embedding code in text
• To embed a simple R calculation within a document \Sexpr
The sum is \Sexpr{1+2}
\Sexpr{paste(“result is”, 2^x)}
Sweave works in a html document
Create a basic html document and process with Sweave
Sweave(“filename.rnw”, driver=RweaveHTML)
<html> <head> <title>Sweave and html</title> </head> <body>
Blah blah <<SweaveCode>>= 1+2 sum(1:10) @ blah blah </body> </html>
Sweave.sty
• Style sheet for R code
RStudio • 4 windows
-Editor, Console, History, Files/plots
• Code completion • Easy access to help (F1) • One step Sweave pdf generation • Searchable history • Keyboard Shortcuts
– http://www.rstudio.org/docs/using/keyboard_shortcuts
• Nice short cut button to build Sweave docs
KnitR
• knitr ≈ Sweave + cacheSweave + pgfSweave + weaver + R2HTML + more
• The design of knitr allows any input languages (e.g. R, Python and Awk) and any output markup languages (e.g. LaTeX, HTML, Markdown and reStructuredText)
• The name knitr was coined with weave in mind, and it also aims to be neater
Features of knitR • Faithful
– knitr writes everything that you see in an R terminal by default (results, plots and warnings)
• Built-in cache • Formatting R code.
– Colors. Uses format R package to “fix code” wrap long lines, add spaces and indent, etc
• Graphics – over 20 graphics devices, can set size etc
• Can use custom regular expressions to parse R
Converting Sweave Rnw to KnitR Rnw
• Very simple • No spaces Chunk names • results='hide' (need quotes) • More chunk options (will review on Rstudio)
Format Source file ending
Output R Code Chunk R expression
Rnw Rnw (.Rnw) Tex, pdf <<R example>>= x <- 1+1rnorm(5) @
\Sexpr{pi}
Github format markdown
Markdown (.Rmd or .md)
md, html ``` {r example} x <- 1+1rnorm(5) ```
`r pi`.
HTML Rhtml .html <!--R example x <- 1+1 rnorm(5) end.rcode-->
<!--rinline pi -->
reStructured Text
.Rst .rst .. {R example} .. x <- 1+1.. rnorm(5) .. .. NOTE:include space after the ..
:r:`pi`
Commands
knit(“tmp.Rnw”)
purl("tmp.Rnw")
knit("example.Rmd")
knit2html("example.Rmd") knit2pdf("example.Rmd")
Markdown using knitR
• Markdown is not latex • Very simple language
eg Emphasis *italic* **bold**
``` {r example} x <- 1+1rnorm(5)
```
```{r} plot(1:10) hist(rnorm(1000)) ```
Versatile – Converting MD with Pandoc • Pandoc a universal document converter
– http://johnmacfarlane.net/pandoc/index.html – Easy to convert markdown file to many formats
pdf file system("pandoc -s example.md -t latex -o
example.pdf")
html file system("pandoc -s example.md -o example.html")
OpenOffice File system("pandoc example.md -o example.odt")
Microsoft Word system("pandoc example.md -o example.docx")
HTML5 Slides
system("pandoc -s -S -i -t dzslides --mathjax slides.md -o slides.html")
http://bcb.dfci.harvard.edu/~aedin/courses/ReproducibleResearch/slides.html
If nothing else…. 1. Organize • Create new folder for each Project
– Can even use Project -> new project in Rstudio
• Store scripts with incremental names – S001project.R, S002project.R etc
• In the top of the folder create a readme text file will list the scripts and what they do
2. Backup • Use a document versioning system
– eg SVN, CVS or GIT. Rstudio has simple support for SVN and GIT
• GIT – load packages directly from GIT into R using the
devtools library function install_github()
• If nothing else store scripts on dropbox or other auto-backup system – So you can revert to previous version if it goes
terribly wrong
3. Make a package
• Easier than you think package.skeleton() • Tutorial on my website
Online publishing - Rpubs
• Free, from Rstudio • Create a new R Markdown Doc
– File -> New -> R Markdown.
• Click the Knit HTML button • Preview click Publish
• http://rpubs.com/
Online Publishing – Shiny
• R package shiny – Shiny allows R developers to build simple
interactive Web-based interfaces for R scripts, using only R code (no JavaScript required!)
– http://www.rstudio.com/shiny/
Please feel free to contact me
aedin@jimmy.harvard.edu
Recommended